Snapshots + Clones

• DSF provides native support (VAAI, ODX, etc)
• Leverage redirect-on-write algorithm
• VM data consists of files (vmdk/vhdx) which are vDisks
• Snapshot taken = vDisk marked immutable
  • New vDisk created as read/write
  • Both vDisks have same block map
  • Metadata mapping to corresponding extents
  • Since each vDisk is its own block map, it eliminates need for snapshot chain

• When VM/vDisk is cloned, current block mapped is locked
• Clones created
• Previously cloned VM acts as “base disk”

• Clones from base VM have their own block map

• New writes/updates occur there

App Consistent Snapshots

• Native VSS for queiscing included
• VmQueisced Snapshot Service (Windows + Linux)

• Post-free: /sbin/pre_freeze
• Post-thaw: /sbin/post_thaw

Eliminating ESXi Stun:

• When delta disks are created, ESXi “stuns” the VM in order to remap disks to new deltas.
• Also occurs when snapshots are deleted.
• During this process, OS cannot execute operations (stuck)
• Duration depends on number of VMDK’s, speed of datastore, etc.
• Nutanix VSS bypasses VMware snapshot/stun process = more efficient

Shadow Clones

• Distributed caching of vDisks/VM data in multi-reader scenario
• Ex. VDI deployment with many linked clones
• Read requests forwarded to “base VM”
• Read requests occur from more than two remote CVM’s (all read I/O) = vDisk marked immutable
• vDisk then cached locally by each VM
• Allows for each node to have a “base VM”